04Proof points

Validated models, and what each was checked against

Ordered roughly by maturity. Every claim below names the experiment or dataset it was tested against.

—Flagship models

Virtual E. coli

K-12 MG1655

What it demonstrates

One integrated kinetic simulation — every process carries a real rate constant, and growth rate emerges from protein accumulation. Built first-principles from biochemistry and enzyme kinetics. ~40,000 reactions/processes; 4,403 genes expressed genome-wide.

Validated against

CRISPRi thyA knockdown/knockout reproduced the epistasis design of Otto et al. 2024, Cell Systems. A live run (Aug 2026): 15 clean doublings, 1 → 32,768 cells. Broader condition-resolved multi-omics validation is in progress, not yet complete.

Virtual A. baumannii

WT + MDR

What it demonstrates

Derived from the E. coli reference. Genetically and chemically perturbable; simulated doubling time ~39 min (lit. ~40 min).

Validated against

EUCAST dose-response across 4 antibiotics. Predicted imipenem effective on WT, ineffective on MDR; sulbactam + durlobactam suppresses MDR growth.

—Also built

Same platform, six more models

Beyond the two flagship models, the same stack has produced:

Dyslipidemia / statin PBPK-PD

Circadian-aware, translating mouse dosing results to human — predicting evening-vs-morning dosing differences.

HIV-1 + T-cell interferon response

36 variables, ported from a published model in ~20 minutes and validated against 3 IFNε conditions from Sazonov et al. 2022, Viruses.

PURE cell-free translation

Reconstructed system matching Matsuura et al. 2017, PNAS Fig. 2B.

SARS-CoV-2

Viral replication, host immune response, and intervention effects in one model.

Virtual immune system

Multiscale, in development — targeting immunotherapy/CAR-T response and toxicity prediction in elderly cancer patients.

Open-source PBPK porting

PK-Sim / OSP Suite mouse-and-human models brought across in ~20 minutes.